Color Image Edges Detection Algorithm Based Upon Anisotropic Gaussian Kernels

Color image edges are the stabilized feature of color image, which have been widely applied in the field of computer vision and image processing. Obtaining accurate positioning and integrity color image edges result is a very important precondition, which can improve the performance of image processing.Comprehensive analysis three kinds of monochromatic-based methods: single channel methods, multidimensional gradient methods and output fusion methods. By experimental simulation and summarizing of advantages and disadvantages of various algorithms, color image edges detection algorithm based upon anisotropic Gaussian kernels(ANGKs) is proposed. The algorithm is divided into two parts: decolorization and edge detection. For first portion, local energy and color contrast based method as well as guided image filter based method are proposed. Color contrast preserving ration, color contrast fidelity ration and harmonic mean of both are used to measure the contrast and luminance preserving condition of proposed decolorization methods. Two methods both can better retain the structure information of color images. For second portion, edge detection algorithm based on ANGKs is proposed. It is white Gaussian noise robust and can describe the directional characteristics of image edges. Non-maximum algorithm and dual threshold selection method are used to deal with the gradient magnitude and direction, which are obtained by Anisotropic Gaussian kernel derivative convolution with images. The algorithm can obtain accurate positioning and white Gaussian noise-robust image edges because multidirectional ANGKs can better describe the directional characteristics of image edges. However, Non-maximum algorithm can cause cross edge fracture. A novel enclosed image edges extraction method is proposed which utilize edge direction and structure. The method can connect the fracture edges.Qualitative method and ROC curves of quantitative method are combined to analysis the whole results of color image edges extraction. The experimental results show that the proposed edge extraction algorithm in the paper has white Gaussian noise robust, richly detailed, accurate positioning and enclosed performance.